Pith. sign in

REVIEW 2 major objections 1 minor 56 references

SFDATrack: Generalized Source-Free Domain Adaptive Tracking Under Adverse Weather Conditions

T0 review · 2 major / 1 minor · reviewed 2026-07-02 · grok-4.3

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SFDATrack adapts visual tracking to adverse weather using only target-domain samples by projecting tokens to multi-domain prototypes.

desk verdict SFDATrack pushes source-free tracking under weather shift with mean-teacher plus new DIM blocks and HPP, but the distillation step from a source-pretrained teacher looks like the unsupported link. read the letter →

arxiv 2607.00369 v1 pith:MRX6BX6E submitted 2026-07-01 cs.CV

Siyuan Yao , Ziqi Wang , Ruiqi Yu , Junqi Huang , Wenqi Ren , Xiaochun Cao This is my paper · ORCID
classification cs.CV
keywords source-freedomainadaptationvisualobjecttrackingadverseweatherconditionshypersphericalprototypeprojectionmean-teacherframeworkmulti-domainprototypesdomain-invariantfeaturesdomain-specificproperties
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper presents SFDATrack as a source-free method for domain-adaptive object tracking that works solely with adverse-weather frames from the target domain. A mean-teacher network with Dual Interactive Mamba blocks first pulls out tokens from augmented samples that remain stable despite weather changes. These tokens are then mapped through a hyperspherical prototype projection module onto a set of prototypes that keep both domain-unique traits and shared invariants. The resulting system produces accurate state estimates across weather types without ever seeing source-domain videos, which are often unavailable under resource limits.

What carries the argument

The hyperspherical prototype projection (HPP) module, which maps distilled tokens onto multi-domain prototypes in latent hyperspherical space while enforcing both domain-specific and domain-invariant properties.

What would settle it

Evaluating SFDATrack on standard adverse-weather tracking benchmarks and finding that accuracy drops below source-dependent baselines when the hyperspherical prototype projection step is removed.

Watch

Extended reading notes

Core claim

SFDATrack first employs a mean-teacher backbone with Dual Interactive Mamba blocks to distill candidate target tokens resilient to weather variations from classified, augmented samples. Afterwards, it introduces a hyperspherical prototype projection module to project these tokens onto multi-domain prototypes within a latent hyperspherical space. By enforcing both domain-specific and domain-invariant properties of the multi-domain prototypes, SFDATrack can be seamlessly adapted to diverse weather conditions with powerful generalizability.

Load-bearing premise

The mean-teacher backbone with Dual Interactive Mamba blocks can distill candidate target tokens that are resilient to weather variations from classified, augmented samples without any source domain data.

Editorial extensions

If this is right

  • The tracker achieves superior performance compared to state-of-the-art approaches on various benchmarks.
  • Adaptation occurs seamlessly across diverse weather conditions while retaining strong generalizability.
  • Robust tracking is possible without access to any source-domain video frames.
  • Only adverse-weather samples from the target domain are required for the adaptation process.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same token-distillation and prototype-enforcement steps could support tracking under other uncontrolled shifts such as lighting or sensor changes.
  • Removing the need for source data may lower privacy and storage barriers when deploying trackers in new locations.
  • A controlled test could measure whether performance remains high when the number of available target-domain samples is reduced.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 1 minor

Summary. The paper proposes SFDATrack, a generalized source-free domain adaptive visual object tracker for adverse weather conditions that uses only adverse weather samples from the target domain. It first applies a mean-teacher backbone incorporating Dual Interactive Mamba (DIM) blocks to distill candidate target tokens resilient to weather variations from classified and augmented samples; these tokens are then projected via a hyperspherical prototype projection (HPP) module onto multi-domain prototypes in a latent hyperspherical space, where both domain-specific and domain-invariant properties are enforced to enable adaptation and generalizability across weather conditions. The abstract asserts that extensive experiments on various benchmarks show superior performance relative to state-of-the-art methods, with code released at a GitHub repository.

Significance. If the central claims hold, the work would represent a meaningful advance in source-free domain adaptation for visual tracking by removing the need for source-domain video data, which is often unavailable due to privacy or storage constraints. The reproducibility strength of releasing code is explicitly noted and supports potential verification and follow-on research in practical adverse-weather tracking scenarios.

major comments (2)
  1. [Abstract] Abstract (second paragraph): The load-bearing claim that the mean-teacher backbone with DIM blocks can distill weather-resilient candidate target tokens solely from classified, augmented target adverse-weather samples is unsupported by any description of pseudo-label generation, teacher initialization handling, or failure-mode analysis; without such a mechanism the source-pretrained features may degrade under unseen conditions, undermining the subsequent HPP module and the overall generalizability assertion.
  2. [Abstract] Abstract (second paragraph): The HPP module is described as enforcing domain-specific and domain-invariant properties on multi-domain prototypes to achieve seamless adaptation, but this step presupposes that the input tokens from the DIM distillation are already resilient; the abstract supplies no empirical check or ablation confirming that the distillation succeeds in the source-free regime, rendering the adaptation claim circular on this unverified precondition.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'classified, augmented samples' is used without specifying the source of the classification labels or the augmentation strategy in a purely target-only, source-free setting.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We agree that the current wording is too concise and does not adequately describe the source-free mechanisms. We will revise the abstract to include brief but explicit references to pseudo-label generation, teacher initialization, and supporting ablations, while preserving the overall length constraints.

read point-by-point responses

  1. Referee: [Abstract] Abstract (second paragraph): The load-bearing claim that the mean-teacher backbone with DIM blocks can distill weather-resilient candidate target tokens solely from classified, augmented target adverse-weather samples is unsupported by any description of pseudo-label generation, teacher initialization handling, or failure-mode analysis; without such a mechanism the source-pretrained features may degrade under unseen conditions, undermining the subsequent HPP module and the overall generalizability assertion.

    Authors: We acknowledge the concern. The abstract currently omits these operational details. In the revised manuscript we will expand the relevant sentence to state that the teacher is initialized from the source-pretrained weights, that pseudo-labels are generated via the teacher’s predictions on weakly augmented target frames, and that a simple consistency filter discards low-confidence tokens. A short failure-mode note will also be added. These clarifications will be supported by the existing algorithmic description and ablation studies in Sections 3.2 and 4.3. revision: yes

  2. Referee: [Abstract] Abstract (second paragraph): The HPP module is described as enforcing domain-specific and domain-invariant properties on multi-domain prototypes to achieve seamless adaptation, but this step presupposes that the input tokens from the DIM distillation are already resilient; the abstract supplies no empirical check or ablation confirming that the distillation succeeds in the source-free regime, rendering the adaptation claim circular on this unverified precondition.

    Authors: We agree that the abstract presents the two stages sequentially without an explicit cross-check. The revised abstract will insert a short clause noting that ablation studies (reported in the supplementary material and Section 4.4) confirm that DIM-distilled tokens retain higher weather robustness than the source-pretrained baseline before HPP projection. This removes the appearance of circularity while keeping the abstract self-contained. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: method pipeline is self-contained with external benchmark validation

full rationale

The provided abstract and description outline a standard source-free adaptation pipeline (mean-teacher + DIM distillation of tokens from target-augmented samples, followed by HPP projection enforcing domain-specific/invariant properties on prototypes). No equations, fitted parameters, or self-citations are shown that reduce any claimed prediction or result to its inputs by construction. Performance is asserted via external benchmarks rather than internal tautology. This matches the default expectation of no significant circularity.

Assumptions & free parameters 0 free parameters · 0 assumptions · 0 invented entities

Abstract-only review prevents enumeration of free parameters, axioms, or invented entities; the two named modules (DIM blocks, HPP) are presented as novel contributions whose internal parameters and assumptions are not described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SFDATrack: Generalized Source-Free Domain Adaptive Tracking Under Adverse Weather Conditions." pith.science (2026). https://pith.science/paper/MRX6BX6E

@misc{pith2026260700369,
  author       = {Pith},
  title        = {Pith review of: SFDATrack: Generalized Source-Free Domain Adaptive Tracking Under Adverse Weather Conditions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MRX6BX6E}},
  note         = {Machine review of arXiv:2607.00369}
}
read the original abstract

Domain adaptive visual object tracking under adverse weather conditions has garnered significant attention in recent years. Despite the impressive performance, existing methods heavily rely on the large-scale video frames from both source and target domains, which is impractical under rigid resource constraints where source data is unavailable. To overcome this limitation, we propose SFDATrack, a generalized source-free domain adaptive tracker that merely leverages adverse weather samples from the target domain for robust state estimation. Specifically, SFDATrack first employs a mean-teacher backbone with Dual Interactive Mamba (DIM) blocks to distill the candidate target tokens that are resilient to weather variations from classified, augmented samples. Afterwards, we introduce a hyperspherical prototype projection (HPP) module to project these tokens onto multi-domain prototypes within a latent hyperspherical space. By enforcing both domain-specific and domain-invariant properties of the multi-domain prototypes, SFDATrack can be seamlessly adapted to diverse weather conditions with powerful generalizability. Extensive experiments evaluated on various benchmarks demonstrate that SFDATrack achieves superior performance compared to state-of-the-art approaches. The code is available at https://github.com/watcherBR0/sfdatrack.

Figures

Figures reproduced from arXiv: 2607.00369 by the authors.

Figure 1
Figure 1. Comparison of the domain adaptive tracking frameworks under adverse weather conditions. (a) Single domain adaptive tracking [50]. (b) Multi-domain adaptive track￾ing [48]. (c) The proposed source-free domain adaptive tracker (SFDATrack). SFDA￾Track adapts a pre-trained model to multiple target domains via latent space projec￾tion, enabling seamless adaptation to diverse weather conditions without requiring any origi… view at source ↗
Figure 2
Figure 2. Overview of the proposed SFDATrack. The input target video is first augmented into weak and strong views, which are processed by the teacher and student encoders, respectively. The extracted features are further fed into the Dual Interactive Mamba (DIM) module and Hyperspherical Prototype Projection (HPP) module to enhance domain consistency and representation robustness. Finally, a total loss L is applied to jointl… view at source ↗
Figure 3
Figure 3. Illustration of the Dual Interactive Mamba (DIM) module. The DIM mod￾ule processes feature tokens from two different augmentations through bidirectional Mamba blocks with prompt tokens exchange. For simplicity, all prime and double￾prime notations (e.g., G′ T , G′′ R) are omitted in the figure. 3.2 Dual Interactive Mamba Although the generic mean-teacher framework provides an efficient way for domain knowledge trans… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visualization results of SFDATrack under three challenging weather conditions: dark, foggy, and rainy scenes. The left part shows the bounding box (BBox) visualiza￾tions, while the right part presents the corresponding heatmaps [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: The performance of our method compared with other trackers in terms of AUC and Precision on eighteen attributes of the real-world AVisT dataset. Each axis corresponds to one attribute, with the radius representing the AUC value under the corresponding condition. Genera…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    In: CVPR

    Bai, Y., Zhao, Z., Gong, Y., Wei, X.: Artrackv2: Prompting autoregressive tracker where to look and how to describe. In: CVPR. pp. 19048–19057 (2024)

    work page 2024

  2. [2]

    In: ECCVW

    Bertinetto, L., Valmadre, J., Henriques, J.F., Vedaldi, A., Torr, P.H.: Fully- convolutional siamese networks for object tracking. In: ECCVW. pp. 850–865 (2016)

    work page 2016

  3. [3]

    In: ICCV

    Bhat, G., Danelljan, M., Gool, L.V., Timofte, R.: Learning discriminative model prediction for tracking. In: ICCV. pp. 6182–6191 (2019)

    work page 2019

  4. [4]

    In: ICCV

    Cai, Y., Liu, J., Tang, J., Wu, G.: Robust object modeling for visual tracking. In: ICCV. pp. 9589–9600 (2023)

    work page 2023

  5. [5]

    In: NeurIPS

    Caron,M., Misra, I., Mairal,J., Goyal, P., Bojanowski, P., Joulin, A.: Unsupervised learning of visual features by contrasting cluster assignments. In: NeurIPS. pp. 9912–9924 (2020)

    work page 2020

  6. [6]

    In: CVPR

    Chen, X., Peng, H., Wang, D., Lu, H., Hu, H.: Seqtrack: Sequence to sequence learning for visual object tracking. In: CVPR. pp. 14572–14581 (2023)

    work page 2023

  7. [7]

    Journal of Image and Graphics27(4), 1191–1203 (2022)

    Chen, Z., Shi, F.: Double template fusion based siamese network for robust visual object tracking. Journal of Image and Graphics27(4), 1191–1203 (2022)

    work page 2022

  8. [8]

    In: ACM KDD

    Chidlovskii, B., Clinchant, S., Csurka, G.: Domain adaptation in the absence of source domain data. In: ACM KDD. pp. 451–460 (2016)

    work page 2016

Show all 56 references
  1. [9]

    In: CVPR

    Cui,Y.,Jiang,C.,Wang,L.,Wu,G.:Mixformer:End-to-endtrackingwithiterative mixed attention. In: CVPR. pp. 13598–13608 (2022)

    2022

  2. [10]

    In: NeurIPS

    Cuturi, M.: Sinkhorn distances: Lightspeed computation of optimal transport. In: NeurIPS. pp. 2292–2300 (2013)

    2013

  3. [11]

    In: ICLR (2021) 16 S

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An image is worth 16x16 words: Transformers for image recognition at scale. In: ICLR (2021) 16 S. Yao et al

    2021

  4. [12]

    In: CVPR

    Fan, H., Lin, L., Yang, F., Chu, P., Deng, G., Yu, S., Bai, H., Xu, Y., Liao, C., Ling, H.: Lasot: A high-quality benchmark for large-scale single object tracking. In: CVPR. pp. 5374–5383 (2019)

    2019

  5. [13]

    In: IROS

    Fu, C., Dong, H., Ye, J., Zheng, G., Li, S., Zhao, J.: Highlightnet: Highlighting low-light potential features for real-time UAV tracking. In: IROS. pp. 12146–12153 (2022)

    2022

  6. [14]

    In: ICARM

    Fu, C., Yao, L., Zuo, H., Zheng, G., Pan, J.: SAM-DA: UAV Tracks Anything at Night with SAM-Powered Domain Adaptation. In: ICARM. pp. 31–38 (2024)

    2024

  7. [15]

    In: WACV

    Gopal, G.Y., Amer, M.A.: Separable self and mixed attention transformers for efficient object tracking. In: WACV. pp. 6708–6717 (2024)

    2024

  8. [16]

    arXiv preprint arXiv:2312.00752 (2023)

    Gu, A., Dao, T.: Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752 (2023)

    2023 arXiv

  9. [17]

    In: CVPR

    Hatamizadeh, A., Kautz, J.: Mambavision: A hybrid mamba-transformer vision backbone. In: CVPR. pp. 25261–25270 (2025)

    2025

  10. [18]

    In: CVPR

    He, A., Luo, C., Tian, X., Zeng, W.: A twofold siamese network for real-time object tracking. In: CVPR. pp. 4834–4843 (2018)

    2018

  11. [19]

    IEEE TPAMI43(5), 1562–1577 (2019)

    Huang, L., Zhao, X., Huang, K.: Got-10k: A large high-diversity benchmark for generic object tracking in the wild. IEEE TPAMI43(5), 1562–1577 (2019)

    2019

  12. [20]

    In: ICLR (2024)

    Hwang, U., Lee, J., Shin, J., Yoon, S.: Sf(da)2: Source-free domain adaptation through the lens of data augmentation. In: ICLR (2024)

    2024

  13. [21]

    In: CVPR

    Kang, B., Zhao, J., Chen, X., Geng, W., Zhang, B., Zhang, L., Wang, D., Lu, H.: Uetrack: A unified and efficient framework for single object tracking. In: CVPR. pp. 20890–20901 (2026)

    2026

  14. [22]

    In: ECCV

    Khanh, T.L.B., Nguyen, H., Pham, L.H., Tran, D.N., Jeon, J.W.: Dynamic retraining-updating mean teacher for source-free object detection. In: ECCV. pp. 328–344 (2024)

    2024

  15. [23]

    In: CVPR

    Li, B., Wu, W., Wang, Q., Zhang, F., Xing, J., Yan, J.: Siamrpn++: Evolution of siamese visual tracking with very deep networks. In: CVPR. pp. 4282–4291 (2019)

    2019

  16. [24]

    In: CVPR

    Li, B., Yan, J., Wu, W., Zhu, Z., Hu, X.: High performance visual tracking with siamese region proposal network. In: CVPR. pp. 8971–8980 (2018)

    2018

  17. [25]

    In: ICRA

    Li, B., Fu, C., Ding, F., Ye, J., Lin, F.: Adtrack: Target-aware dual filter learning for real-time anti-dark UAV tracking. In: ICRA. pp. 496–502 (2021)

    2021

  18. [26]

    IEEE TPAMI46(8), 5743–5762 (2024)

    Li, J., Yu, Z., Du, Z., Zhu, L., Shen, H.T.: A comprehensive survey on source-free domain adaptation. IEEE TPAMI46(8), 5743–5762 (2024)

    2024

  19. [27]

    In: AAAI

    Li,S.,Yeung,D.:Visualobjecttrackingforunmannedaerialvehicles:Abenchmark and new motion models. In: AAAI. pp. 4140–4146 (2017)

    2017

  20. [28]

    In: AAAI

    Li, X., Zhong, B., Liang, Q., Li, G., Mo, Z., Song, S.: Mambalct: Boosting tracking via long-term context state space model. In: AAAI. pp. 4986–4994 (2025)

    2025

  21. [29]

    In: ICML

    Li, Y., Liu, M., Wu, Y., Wang, X., Yang, X., Li, S.: Learning adaptive and view- invariant vision transformer for real-time uav tracking. In: ICML. pp. 28403–28420 (2024)

    2024

  22. [30]

    In: ICML

    Liang, J., Hu, D., Feng, J.: Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation. In: ICML. pp. 6028–6039 (2020)

    2020

  23. [31]

    In: CVPR

    Liu, X., Zhou, L., Zhou, Z., Chen, J., He, Z.: Mambavlt: Time-evolving multimodal state space model for vision-language tracking. In: CVPR. pp. 8731–8741 (2025)

    2025

  24. [32]

    arXiv preprint arXiv:2312.07884 (2023)

    Liu, Y.: Mutual-learning knowledge distillation for nighttime UAV tracking. arXiv preprint arXiv:2312.07884 (2023)

    2023

  25. [33]

    In: NeurIPS

    Mettes, P., van der Pol, E., Snoek, C.: Hyperspherical prototype networks. In: NeurIPS. pp. 1485–1495 (2019) Generalized Source-Free Domain Adaptive Tracking 17

    2019

  26. [34]

    In: BMVC

    Noman, M., Ghallabi, W.A., Kareem, D., Mayer, C., Dudhane, A., Danelljan, M., Cholakkal, H., Khan, S., Gool, L.V., Khan, F.S.: Avist: A benchmark for visual object tracking in adverse visibility. In: BMVC. p. 817 (2022)

    2022

  27. [35]

    In: AAAI

    Shi, L., Zhong, B., Liang, Q., Li, N., Zhang, S., Li, X.: Explicit visual prompts for visual object tracking. In: AAAI. pp. 4838–4846 (2024)

    2024

  28. [36]

    Journal of Image and Graphics27(5), 1493–1508 (2022)

    Shi, Z., Liu, C., Ren, W., Du, S., Zhao, M.: Convolutional neural networks for sand dust image color restoration and visibility enhancement. Journal of Image and Graphics27(5), 1493–1508 (2022)

    2022

  29. [37]

    In: ECCVW

    Varailhon, S., Aminbeidokhti, M., Pedersoli, M., Granger, E.: Source-free domain adaptation for yolo object detection. In: ECCVW. pp. 218–235 (2024)

    2024

  30. [38]

    In: NeurIPS

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. In: NeurIPS. pp. 5998–6008 (2017)

    2017

  31. [39]

    In: AAAI

    Wu, H., Yao, S., Huang, F., Wang, S., Zhang, L., Zheng, Z., Ren, W.: Lvptrack: High performance domain adaptive UAV tracking with label aligned visual prompt tuning. In: AAAI. pp. 8395–8403 (2025)

    2025

  32. [40]

    In: CVPR

    Wu, Q., Yang, T., Liu, Z., Wu, B., Shan, Y., Chan, A.B.: Dropmae: Masked au- toencoders with spatial-attention dropout for tracking tasks. In: CVPR. pp. 14561– 14571 (2023)

    2023

  33. [41]

    In: CVPR

    Wu, Y., Wang, X., Yang, X., Liu, M., Zeng, D., Ye, H., Li, S.: Learning occlusion- robust vision transformers for real-time uav tracking. In: CVPR. pp. 17103–17113 (2025)

    2025

  34. [42]

    In: CVPR

    Xie, J., Zhong, B., Mo, Z., Zhang, S., Shi, L., Song, S., Ji, R.: Autoregressive queries for adaptive tracking with spatio-temporal transformers. In: CVPR. pp. 19300–19309 (2024)

    2024

  35. [43]

    In: CVPR

    Xue, C., Zhong, B., Liang, Q., Zheng, Y., Li, N., Xue, Y., Song, S.: Similarity- guided layer-adaptive vision transformer for uav tracking. In: CVPR. pp. 6730– 6740 (2025)

    2025

  36. [44]

    In: ICCV

    Yan, B., Peng, H., Fu, J., Wang, D., Lu, H.: Learning spatio-temporal transformer for visual tracking. In: ICCV. pp. 10428–10437 (2021)

    2021

  37. [45]

    IEEE TIP34, 3533– 3546 (2025)

    Yao, S., Guo, Y., Yan, Y., Ren, W., Cao, X.: Unctrack: Reliable visual object tracking with uncertainty-aware prototype memory network. IEEE TIP34, 3533– 3546 (2025)

    2025

  38. [46]

    IEEE TIP30, 4814–4827 (2021)

    Yao,S.,Han,X.,Zhang,H.,Wang,X.,Cao,X.:Learningdeeplucas-kanadesiamese network for visual tracking. IEEE TIP30, 4814–4827 (2021)

    2021

  39. [47]

    IEEE TIP33, 5936–5948 (2024)

    Yao, S., Sun, H., Xiang, T., Wang, X., Cao, X.: Hierarchical graph interaction transformer with dynamic token clustering for camouflaged object detection. IEEE TIP33, 5936–5948 (2024)

    2024

  40. [48]

    In: ICCV

    Yao, S., Zhu, R., Wang, Z., Ren, W., Yan, Y., Cao, X.: Umdatrack: Unified multi- domain adaptive tracking under adverse weather conditions. In: ICCV. pp. 6466– 6475 (2025)

    2025

  41. [49]

    In: ECCV

    Ye, B., Chang, H., Ma, B., Shan, S., Chen, X.: Joint feature learning and relation modeling for tracking: A one-stream framework. In: ECCV. pp. 341–357 (2022)

    2022

  42. [50]

    In: IROS

    Ye, J., Fu, C., Zheng, G., Cao, Z., Li, B.: Darklighter: Light up the darkness for UAV tracking. In: IROS. pp. 3079–3085 (2021)

    2021

  43. [51]

    In: CVPR

    Ye, J., Fu, C., Zheng, G., Paudel, D.P., Chen, G.: Unsupervised domain adaptation for nighttime aerial tracking. In: CVPR. pp. 8886–8895 (2022)

    2022

  44. [52]

    In: ECCV

    Yoon, I., Kwon, H., Kim, J., Park, J., Jang, H., Sohn, K.: Enhancing source-free domain adaptive object detection with low-confidence pseudo label distillation. In: ECCV. pp. 337–353 (2024)

    2024

  45. [53]

    arXiv preprint arXiv:2106.07862 (2021) 18 S

    Zhang, Z., Zhang, L.: Domain adaptive siamrpn++ for object tracking in the wild. arXiv preprint arXiv:2106.07862 (2021) 18 S. Yao et al

    2021

  46. [54]

    In: AAAI

    Zheng, Y., Zhong, B., Liang, Q., Mo, Z., Zhang, S., Li, X.: Odtrack: Online dense temporal token learning for visual tracking. In: AAAI. pp. 7588–7596 (2024)

    2024

  47. [55]

    In: ICRA

    Zhu,J.,Tang,H.,Cheng,Z.,He,J.,Luo,B.,Qiu,S.,Li,S.,Lu,H.:DCPT:darkness clue-prompted tracking in nighttime uavs. In: ICRA. pp. 7381–7388 (2024)

    2024

  48. [56]

    In: ICML

    Zhu, L., Liao, B., Zhang, Q., Wang, X., Liu, W., Wang, X.: Vision mamba: Efficient visual representation learning with bidirectional state space model. In: ICML. pp. 62429–62442 (2024)

    2024

Pith tools

Reviewed July 2, 2026 · model on record in the stance chip above.